Machine learning techniques and AI models are proving useful across many application domains. However, the application of AI in computer networking remains challenging. Just as networking is an integral part of modern computing systems, the usefulness of machine learning techniques and AI models in this domain depends upon the ability to access, share and use good data. Data is needed to train and refine models for the ultimate implementation of good models in networks. However, data remains siloed and in many formats across institutions and in industry..
Advances in AI techniques can enhance performance and security in computer and networking systems. Significant and impactful efforts are emerging across public and private sectors to advance AI research and development. The use of AI techniques in cybersecurity - in malware and phishing detection, for example - has been incorporated into mainstream tools such as endpoint detection and spam filters. But there remains a lack of focus on topics at the intersection of networking and AI. Networks and their associated data are notoriously inaccessible to researchers. This is in part due to a lack of available resources and infrastructure to collect and support such research, including data, testbeds, and benchmarks, as well as the proprietary nature of networks, especially in the commercial space.
This two-day NSF-funded workshop sought to explore the fundamental needs that underpin the uses of AI for networking, including topics such as: what data can be made available for AI-enabled network systems? How will network data be collected, curated and used? What properties of data, and in what ways, would impact the networks and applications from a technical, legal, and ethical context? What new testbeds, labeling techniques, benchmarks, and benchmarking techniques are needed for network applications? And, most importantly, how will both the mindsets and skillsets of network researchers and our next generation of students evolve and transform as we head into the future?